UA-Bench
Models identify underwater sound sources and their subclasses from audio recordings. The authors repeat the test with silent audio to check reliance on sound.
- Made by
- Zhe Sun and colleagues
- Released
- August 2026
- Kind
- Audio understanding and voice
- Level
- University
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Multi-turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Audio and speech
- Last model added
- June 2026
- Scores refreshed
- November 20th, 2026
Qwen2.5-Omni-7B leads with 13.1% among models released by June 2026 Qwen2.5-Omni-7B leads with 13.1% among models released by June 2026
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in June 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1
Qwen2.5-Omni-7B Strict single-letter decoder, zero-shot 13.1%
- 2 MiDashengLM-7B-1021 Strict single-letter decoder, zero-shot 10.3%
- 3 Audio-Flamingo-Next Strict single-letter decoder, zero-shot 8.10%
- 4
Voxtral Small Strict single-letter decoder, zero-shot 7.70%
- 5
Qwen2-Audio-7B-Instruct Strict single-letter decoder, zero-shot 7.30%
- 6 Aero-1-Audio Strict single-letter decoder, zero-shot 6.90%
- 7
Gemma 4 E2B Strict single-letter decoder, zero-shot 6.50%
- 8 Voxtral-Mini-3B-2507 Strict single-letter decoder, zero-shot 6.10%
- 9
Gemma 4 12B Strict single-letter decoder, zero-shot 2.00%
Source: The UA-Bench paper
- 1
Qwen2.5-Omni-7B Strict single-letter decoder, zero-shot 13.1%
- 2
Voxtral Small Strict single-letter decoder, zero-shot 7.70%
- 3
Qwen2-Audio-7B-Instruct Strict single-letter decoder, zero-shot 7.30%
- 4
Gemma 4 E2B Strict single-letter decoder, zero-shot 6.50%
- 5
Gemma 4 12B Strict single-letter decoder, zero-shot 2.00%
Source: The UA-Bench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 13.1% | Strict single-letter decoder, zero-shot | Mar 2025 | Official leaderboard |
| 2 | MiDashengLM-7B-1021 | 10.3% | Strict single-letter decoder, zero-shot | Official leaderboard | |
| 3 | Audio-Flamingo-Next | 8.10% | Strict single-letter decoder, zero-shot | Official leaderboard | |
| 4 | | 7.70% | Strict single-letter decoder, zero-shot | Jul 2025 | Official leaderboard |
| 5 | | 7.30% | Strict single-letter decoder, zero-shot | Aug 2024 | Official leaderboard |
| 6 | Aero-1-Audio | 6.90% | Strict single-letter decoder, zero-shot | Official leaderboard | |
| 7 | | 6.50% | Strict single-letter decoder, zero-shot | Mar 2026 | Official leaderboard |
| 8 | Voxtral-Mini-3B-2507 | 6.10% | Strict single-letter decoder, zero-shot | Official leaderboard | |
| 9 | | 2.00% | Strict single-letter decoder, zero-shot | Jun 2026 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 13.1% | Strict single-letter decoder, zero-shot | Mar 2025 | Official leaderboard |
| 2 | | 7.70% | Strict single-letter decoder, zero-shot | Jul 2025 | Official leaderboard |
| 3 | | 7.30% | Strict single-letter decoder, zero-shot | Aug 2024 | Official leaderboard |
| 4 | | 6.50% | Strict single-letter decoder, zero-shot | Mar 2026 | Official leaderboard |
| 5 | | 2.00% | Strict single-letter decoder, zero-shot | Jun 2026 | Official leaderboard |
Rankings by part
We read 3 parts from the benchmark’s paper: Detection pulse, Communication waveform and Ship-radiated noise. Each ranks models on one part of the test. Voxtral Small leads 1 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Detection pulse
5 models we track have a score here.
- 1
Voxtral Small ↑1 23.2%
- 2
Qwen2.5-Omni-7B ↓1 15%
- 3
Gemma 4 12B ↑2 4.50%
- 4
Gemma 4 E2B 4.30%
- 5
Qwen2-Audio-7B-Instruct ↓2 0%
Communication waveform
5 models we track have a score here.
- 1
Gemma 4 E2B ↑3 12%
- 2
Qwen2.5-Omni-7B ↓1 5.30%
- 3
Voxtral Small ↓1 1.40%
- 4
Gemma 4 12B ↑1 1.00%
- 5
Qwen2-Audio-7B-Instruct ↓2 0%
Ship-radiated noise
5 models we track have a score here.
- 1
Qwen2.5-Omni-7B 19.8%
- 2
Qwen2-Audio-7B-Instruct ↑1 19%
- 3
Voxtral Small ↓1 4.80%
- 4
Gemma 4 E2B 2.30%
- 5
Gemma 4 12B 1.40%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The UA-Bench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official UA-Bench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Audio and video
- Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy.
Qwen 3.8 Max 88.7%
- LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization.
Gemini 4 Argon 91.7%
- MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy.
Qwen 3.8 Max 82.4%
- MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy.
Qwen 3.8 Max 90.8%
- FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better.
Phi-4-multimodal 4.00%
- VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules.
Qwen3.5-Omni-Plus 93.1%
Data audit
All 10 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (9 entries).
- Pass Models scored 5 models we track, plus 4 other entries on its leaderboard.
- Pass Newest score The newest is from June 2026, for Gemma 4 12B.
- Note Frontier models None of its models are on the capabilities index.
- Pass Headroom The best is 13.1%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on sunzhe12.github.io: 5 rows for models we track and 4 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.